AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Temperature in Cosine-based Softmax Loss

Takumi Kobayashi

ICCV 2025 · Label-efficient and robust learning › Recognition and long-tail · approach: Loss design and classifier theory · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use least-squares kappa instead of rule-based kappacosine softmax ResNet-50 · ImageNetaccuracy: 3.36 -> 77.27Improved
use least-squares kappa instead of maximum-likelihood kappacosine softmax ResNet-50 · ImageNetaccuracy: 0.10 -> 77.27Improved
use LS kappa instead of trainable kappacosine softmax ResNet-50 · ImageNetaccuracy: 75.51 -> 77.27Improved
set kappa to 2kappa* instead of softmaxResNet-50 · ImageNetaccuracy: 77.12 -> 77.81Improved
raise fixed kappa from 5 to 20 (ablation)cosine softmax ResNet-50 · ImageNetaccuracy: 68.95 -> 77.76Improved
use kappa=60 instead of kappa* (ablation)cosine softmax · ImageNetaccuracy: 77.27 -> 76.88Hurt
larger kappa=60 vs softmax (ablation)ResNet-50 pretrained · CUB-200 linear probeaccuracy: 69.05 -> 75.63Improved
larger kappa=60 vs softmax (ablation)ResNet-50 pretrained · Flower-102 linear probeaccuracy: 86.39 -> 92.19Improved
larger kappa=60 vs softmax (ablation)ResNet-50 pretrained · CUB-200 fine-tuningaccuracy: 80.17 -> 81.18Improved
small kappa=10 vs softmax (ablation)ResNet-50 pretrained · CUB-200 linear probeaccuracy: 69.05 -> 47.13Hurt
use LS kappa in long-tailed trainingResNet-50 · iNaturalist2018accuracy: 66.76 -> 67.75Improved
use fixed kappa=10 instead of LS kappa*=18.54 (ablation)ResNet-50 · iNaturalist2018accuracy: 67.75 -> 57.36Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy